IEEE Sensors Journal最新文献

{"title":"A Self-Supervised Framework Based on Time-Frequency Contrastive Learning for Fault Diagnosis With Few Labeled Data","authors":"Zong Meng;Zengjin Yan;Haoze Chen;Jimeng Li;Xiyuan Zhang;Fengjie Fan","doi":"10.1109/JSEN.2026.3715802","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3715802","url":null,"abstract":"In practical engineering applications, labeled data collected from bearings are often scarce. The majority of intelligent diagnostic methods encounter challenges in identifying effective sample relationships in the absence of sufficient labeled samples. In addressing the challenge posed by the scarcity of labeled data, this article puts forth a time-frequency self-supervised contrastive learning (TFSSCL) method. This technique builds self-contrastive by using the time-domain and frequency-domain features of vibration signals, learning unsupervised features by maximizing the similarity of time-frequency characteristics, and reducing dependence on labeled samples and negative samples. First, the time-domain signals enhanced with data augmentation are compared with the predicted frequency-domain signals. Second, a time-frequency feature extraction model incorporating a temporal convolutional network (TCN) and an attention module tailored for time-frequency signals is constructed to measure the correlation between the time-frequency signals of unlabeled samples. Finally, a joint loss function based on a cross-correlation matrix and time-frequency self-contrastive is constructed to enhance the network’s discriminating ability. Diagnostic results show that the TFSSCL model maintains good classification accuracy with limited labeled data.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26477-26484"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871462","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Flexible LCR Sensor Based on Heterogeneous Composite Electrodes and Active Loss Modulation","authors":"Zihao Li;Shixiong Song;Jing Miao;Yingping Hong;Boshan Sun;Jijun Xiong;Chen Li","doi":"10.1109/JSEN.2026.3714583","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3714583","url":null,"abstract":"Flexible wireless passive sensors often suffer from cross-interference caused by temperature drifts and spatial fluctuations in complex environments. While conventional <inline-formula> <tex-math>$LC$ </tex-math></inline-formula> sensors are inherently limited by single-parameter output frameworks that rely solely on resonant frequency tracking, the exploration of additional sensing dimensions remains challenging due to the strict requirement of maintaining a high quality factor (<italic>Q</i>). To break this constraint and expand the observation window, this article proposes an <inline-formula> <tex-math>$LCR$ </tex-math></inline-formula> sensing architecture based on heterogeneous composite electrodes and an active loss modulation strategy. By leveraging a high-Q spiral inductor to compensate for initial dissipation, the architecture integrates a silver-paste-based stable geometric capacitance and a liquid metal eutectic gallium–indium (EGaIn)-based strain-sensitive resistor on a thermoplastic polyurethane (TPU) substrate. Operating in the inherent under-coupled regime of noncontact measurements, the sensor exploits the strain-induced exacerbation of impedance mismatch to translate mechanical strain into a dual-parameter orthogonal response: resonant frequency shift and return loss (<inline-formula> <tex-math>$textit {S}_{{11}}$ </tex-math></inline-formula>) attenuation. Experimental results demonstrate that the system exhibits precise feature isolation over a 0%–50% tensile strain range. Specifically, the <inline-formula> <tex-math>$textit {S}_{{11}}$ </tex-math></inline-formula> amplitude fluctuation is suppressed to < 1 dB across a wide temperature range (<inline-formula> <tex-math>$25~^{circ }$ </tex-math></inline-formula>C–<inline-formula> <tex-math>$100~^{circ }$ </tex-math></inline-formula>C), while the maximum drift of <inline-formula> <tex-math>$textit {f}_{{0}}$ </tex-math></inline-formula> is merely 0.5% under coupling distance variations from 5 to 15 mm. Enabled by these orthogonal characteristics, the system effectively eliminates multiphysics signal crosstalk, providing a viable device-level strategy for highly robust monitoring in complex environments.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"25504-25513"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871010","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Highly Sensitive Flexible Battery-Type Self-Powered Pressure Sensor With a Wide Detection Range","authors":"Shaobin Liu;Qiang Wang;Yalong Yang;Liangliang Su;Xulai Zhu;Zhu Ma;Wei Zeng","doi":"10.1109/JSEN.2026.3715661","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3715661","url":null,"abstract":"Flexible self-powered pressure sensors hold considerable promise for wearable electronics, health monitoring, and human–machine interfaces, yet it remains challenging to simultaneously achieve high sensitivity, a wide detection range, and stable output. Here, we present a flexible battery-type self-powered pressure sensor with high sensitivity and a wide detection range. A polyurethane (PU) foam spacer is introduced to regulate the effective contact between the Zn foil and the PAM-PU-HG hydrated hydrogel layer. Under external pressure, the PU foam undergoes reversible compression, which enhances interfacial contact and reduces the apparent internal resistance of the device, thereby generating a pressure-dependent voltage output without an external power supply. The fabricated sensor exhibits a sensitivity of 45.8 mV kPa^–1 in the low-pressure range of 0–40 kPa and a detection range up to 338 kPa, with response and recovery times of 169 and 204 ms, respectively, while maintaining stable output over 1750 repeated loading/unloading cycles. The device further enables the monitoring of various human motions. By combining Morse coding with a CNN-BiGRU-Attention model, automatic recognition of 26 alphabet classes was achieved with an accuracy of 99.62%. This work provides a viable route for the structural design of high-performance battery-type self-powered pressure sensors and for their application in human–machine interfaces.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"25514-25520"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"sEMG-Based Force Estimation With Channel Reduction for Human--Robot Interaction in Forearm Rehabilitation: Online Experimental Validation","authors":"Thantip Sittiruk;Kiattisak Sengchuai;Apidet Booranawong;Pornchai Phukpattaranont;Hiroshi Saito","doi":"10.1109/JSEN.2026.3709760","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3709760","url":null,"abstract":"Surface electromyography (sEMG)-based human force estimation has been widely applied in prosthetic control, rehabilitation assistance, and human–machine interaction systems. This study investigates forearm muscle force estimation under different experimental conditions involving multiple elbow placement patterns that represent natural movement trajectories in <italic>XY</i>-planar rehabilitation mobile robots. A Gaussian process regression GPR) model with an exponential kernel was identified as the optimal regression approach and implemented in Simulink to estimate forces from eight-channel sEMG signals. In addition, a root mean square error (RMSE)-based channel reduction method for <italic>X</i>- and <italic>Y</i>-direction force estimation across four movement directions is proposed to reduce the number of required channels from the full Myo armband configuration. Experimental results show that five optimal channels achieved an average <italic>XY</i> force estimation error of 1.860 N in offline evaluation. In online testing, the average estimation error increased by 3.898 N, approximately doubling the overall estimation error compared with the offline results.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26618-26633"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871101","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Hybrid CNN–BiLSTM Framework With Quantum-Inspired Attention Mechanism for Underwater Acoustic Signal Classification","authors":"Mintu Kumar;Neel Kanth Kundu","doi":"10.1109/JSEN.2026.3713248","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3713248","url":null,"abstract":"The classification of underwater acoustic signals is essential for maritime surveillance, naval operations, and underwater target identification. However, this task is challenging due to environmental noise and variability in acoustic signals. The existing deep learning models are often complex and computationally intensive, limiting their practicality in real underwater situations. In this work, we propose a lightweight and effective hybrid model that combines a convolutional neural network–bidirectional long short-term memory (CNN–BiLSTM) architecture with a quantum-inspired attention (QIA) module. Time–frequency features of the input signals are extracted using the Mel spectrogram, gammatonegram, and constant-Q transform (CQT) and fused to form a three-channel time–frequency representation. This representation is passed through a CNN module for spatial feature extraction, followed by a BiLSTM module for temporal modeling, and a final QIA module to enhance discriminative focus. The proposed model was evaluated on the vessel-type underwater acoustic data (VTUAD) and achieved notable classification performance across multiple range-based scenarios. The proposed model outperformed the previous baseline ResNet18 model, with improvements of 1.13% in the 2-km inclusion and 4-km exclusion scenario, 3.04% in the 3-km inclusion and 5-km exclusion scenario, 3.31% in the 4-km inclusion and 6-km exclusion scenario, and 12.59% in the combined scenario while using only 0.47 M parameters corresponding to a 95.8% reduction in model parameters. These results indicate the model’s robustness to noise and its stable performance under high signal-to-noise ratio (SNR) variability in combined-distance scenarios. To the best of our knowledge, this work represents one of the earliest studies exploring quantum-inspired machine learning (QIML) for underwater acoustic signal classification (UASC), demonstrating its potential for high-performance and resource-efficient acoustic sensing in complex underwater environments.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26218-26230"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871156","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Robust Data-Driven Method for Data and Label Noise Mitigation in Synchronous Machine Winding Short-Circuit Fault Diagnosis","authors":"Yu Chen;Yuxuan Ding;Qianchao Wang;Chakhung Yeung;Chenguo Yao;Yaping Du;Zhongyong Zhao","doi":"10.1109/JSEN.2026.3713996","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3713996","url":null,"abstract":"Winding short-circuit (SC) faults pose a critical challenge in synchronous machines, where accurate and timely diagnosis is essential to ensure operational reliability and prevent cascading failures in the power system. Conventional inspection of synchronous machine windings relies heavily on labor-intensive expert assessment, which is often subjective and inefficient. Although recent data-driven approaches have advanced SC fault diagnosis, they require training samples with low noise and accurate labels, incurring prohibitive costs. To address these limitations, this study proposes a robust data-driven method for joint mitigation of data and label noise in synchronous machine winding SC fault diagnosis. To mitigate data noise, binary morphology is used to preprocess frequency response analysis (FRA) data before input into the data-driven model, extracting informative frequency bands while suppressing perturbations. For label noise, an auxiliary model combined with the small-loss criterion identifies mislabeled samples in the training dataset, followed by virtual adversarial training (VAT) to utilize all samples, treating mislabeled samples as unlabeled, in a semisupervised learning mode. The proposed method is validated on 5- and 7.5- kVA synchronous machine FRA datasets. Under the controlled combined noise settings, it maintains accuracies above 90% even when the training data contain up to 40% incorrect labels and additive Gaussian data noise down to 10-dB signal-to-noise ratio (SNR), with the remaining misclassifications mainly occurring among inherently similar fault categories.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26294-26306"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871188","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Time-Frequency Aware Residual GAN Based on Dynamic Physical Models for Bearing Digital Twin Modeling","authors":"Qindan Luo;Renxiang Chen;Yi Qin;Yu Huang;Chenghao Li","doi":"10.1109/JSEN.2026.3709338","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3709338","url":null,"abstract":"Addressing the issue of insufficient fidelity in bearing digital twins (DTs) during full-lifecycle modeling, generating high-fidelity and interpretable twin data is of significant research importance. Therefore, this article proposes a method for establishing a full-lifecycle bearing DT model based on a dynamic physics-informed time-frequency perception residual generative adversarial network (TFPRGAN). First, a mapping mechanism between the DT and the physical bearing is constructed in the virtual space. By establishing a bearing dynamics model in virtual space to output twin signals, the optimization objective is set as the impact resonance characteristics caused by actual bearing defects. These signals are then iteratively refined through the proposed TFPRGAN. In the network, dilated convolutions are employed in the temporal path to capture multiscale features and expand the receptive field, while a residual network is introduced in the frequency path to extract energy features from fault-bearing images obtained via short-time Fourier transform (STFT). The loss function is further improved to minimize the distribution divergence between the twin signals and the measured signals. The fidelity of the refined signals is evaluated using root mean square (rms)-based defect-size prediction and envelope analysis indicators. Experiments on a full-lifecycle bearing dataset show that the twin model refined by TFPRGAN achieves higher fidelity compared to other model-correction methods, validating the effectiveness of the proposed approach.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"25916-25927"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Piezoresistive Compressive Strain Sensor Based on MWCNT-Coated Porous PDMS Foam: Fabrication, Characterization, and Wearable Application","authors":"Abhinav Sharma;Mohd. Zahid Ansari","doi":"10.1109/JSEN.2026.3716800","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3716800","url":null,"abstract":"Flexible piezoresistive strain sensors capable of reliable compressive strain detection are increasingly in demand for applications in wearable health monitoring, soft robotics, and human–machine interfaces. In this work, a compressive piezoresistive strain sensor based on multiwalled carbon nanotube (MWCNT)-coated porous polydimethylsiloxane (PDMS) foam is fabricated and characterized. The porous PDMS foam was prepared via sacrificial sugar-cube templating and functionalized with a MWCNT conductive network through dip coating. A custom compression-testing setup was developed to enable controlled loading and real-time electrical measurements, ensuring reliable evaluation of the sensor’s electromechanical response. The fabricated sensor exhibits a baseline resistance of 13 k<inline-formula> <tex-math>$Omega $ </tex-math></inline-formula> and demonstrates a monotonically increasing relative resistance change (<inline-formula> <tex-math>$Delta textit {R}/textit {R}_{{0}}$ </tex-math></inline-formula>) across a compressive strain range of 0%–70%, with a maximum gauge factor (GF) of approximately 4 in the low-strain regime (0%–20%). Dynamic testing reveals a response time of 0.48 s and a recovery time of 0.60 s, along with an independent electrical output across loading rates of 5–40 mm <inline-formula> <tex-math>$cdot $ </tex-math></inline-formula> <inline-formula> <tex-math>${mathrm{min}}^{-{1}}$ </tex-math></inline-formula>. Long-term cyclic durability testing over 2000 compression–release cycles at 30% strain demonstrates stable and repeatable <inline-formula> <tex-math>$Delta textit {R}/textit {R}_{{0}}$ </tex-math></inline-formula> output with limited baseline drift. The practical utility of the sensor is demonstrated through proof-of-concept experiments: finger-click detection, continuous elbow-bending motion monitoring, and real-time compression level classification via an ESP32-based wireless platform. These results show the MWCNT-coated porous PDMS foam as a practical, simple, and potentially scalable platform for flexible compressive strain sensing in wearable and embedded sensing systems.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"25530-25540"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871226","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Node Coverage Optimization for Field Observation Instrument Networks Based on Virtual Force-Guided Slime Mold Optimization Algorithm","authors":"Fang Cao;Jiuyuan Huo;Jiguang Yang;Shannong Zheng","doi":"10.1109/JSEN.2026.3709776","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3709776","url":null,"abstract":"Ensuring comprehensive coverage in cold and arid regions remains a critical challenge in wireless sensor network (WSN) deployment. To address this issue, this article proposes a coverage optimization method based on a probabilistic sensing model and a virtual force-guided slime mold optimization algorithm for the deployment of nodes in field observation instrument networks (FOI-VFSMA). The proposed approach introduces several key innovations. First, mobile node positions are initialized using the Sobol low-discrepancy sequence to enhance the initial coverage rate. Second, coverage, distribution uniformity, and node movement distance are jointly defined as optimization objectives, and their weights are dynamically adjusted using the entropy weight method to balance competing goals. Third, during node position updates, the slime mold algorithm is integrated with a virtual force model to introduce position perturbations, thereby avoiding local optima and enhancing global search ability. Furthermore, heterogeneous nodes with differentiated sensing capabilities are incorporated to improve network adaptability in complex environments. Extensive experiments are conducted with twelve scenarios containing multiple types of obstacles, while also considering mobile obstacles and node failures. The results demonstrate that FOI-VFSMA consistently outperforms comparative algorithms in terms of coverage, uniformity, and average movement distance, maintaining strong robustness and adaptability even under complex and dynamic conditions.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26838-26850"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871236","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MARVEL: A Self-Supervised Masked Autoencoder for Robust Multibeam DVL Velocity Reconstruction in Aquatic Navigation","authors":"Jianan Lou;Shuke Wang;Shaolin Lü;Jun Yang;Bo Hou;Rong Zhang","doi":"10.1109/JSEN.2026.3714089","DOIUrl":"https://doi.org/10.1109/JSEN.2026.3714089","url":null,"abstract":"Multibeam Doppler velocity logs (DVLs) are a cornerstone sensor in aquatic navigation, yet individual beams are frequently corrupted or lost in practice, leading to partial observations and degraded performance. Existing nonlearning approaches rely on heuristic beam screening and hand-crafted models, while prior learning-based methods often require training separate networks for different missing-beam patterns. This article formulates DVL beam loss and degradation as a beam-velocity reconstruction problem and proposes MARVEL, a self-supervised masked autoencoder for completing missing along-beam velocities from partially observed multibeam measurements. MARVEL is trained on complete DVL data with realistic synthetic beam masking, enabling a single unified model to handle arbitrary missing-beam patterns at deployment. Experiments on multiple real-world aquatic datasets demonstrate that MARVEL consistently reduces beam-velocity reconstruction error compared with conventional heuristic baselines and prior learning-based approaches. Efficiency analyses further show that MARVEL provides a practical, single-model solution with reduced training overhead and lightweight inference, facilitating real-time deployment.","PeriodicalId":447,"journal":{"name":"IEEE Sensors Journal","volume":"26 17","pages":"26904-26917"},"PeriodicalIF":4.5,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148871243","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}